A Novel Model for Predicting PM2.5 Concentrations Utilizing Graph Convolutional Networks and Transformer

With the development of the global economy, PM2.5 fine particulate matter concentration has emerged as a major environmental issue worldwide, significantly impacting human health. However, most existing research methods largely ignore the spatial characteristics of PM2.5 concentrations. In response,...

Full description

Saved in:
Bibliographic Details
Main Authors: Yuan Huang, Feilong Han, Qimeng Feng
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10884736/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849720413578330112
author Yuan Huang
Feilong Han
Qimeng Feng
author_facet Yuan Huang
Feilong Han
Qimeng Feng
author_sort Yuan Huang
collection DOAJ
description With the development of the global economy, PM2.5 fine particulate matter concentration has emerged as a major environmental issue worldwide, significantly impacting human health. However, most existing research methods largely ignore the spatial characteristics of PM2.5 concentrations. In response, this paper proposes a new approach based on Graph Convolutional Networks (GCN) and Transformer. To enhance the model’s predictive performance, we designed a new Transformer architecture named FFPformer, which incorporates the Fast Fourier Transform into the Transformer framework. Initially, adjacency matrices are constructed using geographic latitude, longitude, and altitude information to represent the spatial relationships among monitoring stations. These spatial relationships are then extracted using GCN. The extracted spatial features are transformed into value, position, and temporal representations via embedding blocks. The encoded temporal information is converted into frequency domain representations through the encoding layer, then reconverted into temporal information after attention calculations, and input into the decoding layer to produce the prediction results. Finally, a Huber loss is used to optimize the neural network parameters and enhance the robustness of the model. The GCN-FFPformer model has been compared with traditional time series models and advanced Transformer models using real-world datasets. The results indicate that on the Beijing-Tianjin-Hebei dataset, MAE, RMSE, and TIC were reduced by 9.65%, 11.67%, and 12.51% on average compared to other models, demonstrating that GCN-FFPformer is a novel method for accurately predicting PM2.5 concentrations in urban areas.
format Article
id doaj-art-9469f9932bcd4ac9930ff2c5bc062519
institution DOAJ
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-9469f9932bcd4ac9930ff2c5bc0625192025-08-20T03:11:55ZengIEEEIEEE Access2169-35362025-01-0113306133062210.1109/ACCESS.2025.354177410884736A Novel Model for Predicting PM2.5 Concentrations Utilizing Graph Convolutional Networks and TransformerYuan Huang0Feilong Han1https://orcid.org/0009-0002-6533-2519Qimeng Feng2School of Information and Electrical Engineering, Hebei University of Engineering, Handan, ChinaSchool of Information and Electrical Engineering, Hebei University of Engineering, Handan, ChinaSchool of Information and Electrical Engineering, Hebei University of Engineering, Handan, ChinaWith the development of the global economy, PM2.5 fine particulate matter concentration has emerged as a major environmental issue worldwide, significantly impacting human health. However, most existing research methods largely ignore the spatial characteristics of PM2.5 concentrations. In response, this paper proposes a new approach based on Graph Convolutional Networks (GCN) and Transformer. To enhance the model’s predictive performance, we designed a new Transformer architecture named FFPformer, which incorporates the Fast Fourier Transform into the Transformer framework. Initially, adjacency matrices are constructed using geographic latitude, longitude, and altitude information to represent the spatial relationships among monitoring stations. These spatial relationships are then extracted using GCN. The extracted spatial features are transformed into value, position, and temporal representations via embedding blocks. The encoded temporal information is converted into frequency domain representations through the encoding layer, then reconverted into temporal information after attention calculations, and input into the decoding layer to produce the prediction results. Finally, a Huber loss is used to optimize the neural network parameters and enhance the robustness of the model. The GCN-FFPformer model has been compared with traditional time series models and advanced Transformer models using real-world datasets. The results indicate that on the Beijing-Tianjin-Hebei dataset, MAE, RMSE, and TIC were reduced by 9.65%, 11.67%, and 12.51% on average compared to other models, demonstrating that GCN-FFPformer is a novel method for accurately predicting PM2.5 concentrations in urban areas.https://ieeexplore.ieee.org/document/10884736/PM2.5 concentration predictionspatiotemporal predictiongraph convolutional networks (GCN)Transformer
spellingShingle Yuan Huang
Feilong Han
Qimeng Feng
A Novel Model for Predicting PM2.5 Concentrations Utilizing Graph Convolutional Networks and Transformer
IEEE Access
PM2.5 concentration prediction
spatiotemporal prediction
graph convolutional networks (GCN)
Transformer
title A Novel Model for Predicting PM2.5 Concentrations Utilizing Graph Convolutional Networks and Transformer
title_full A Novel Model for Predicting PM2.5 Concentrations Utilizing Graph Convolutional Networks and Transformer
title_fullStr A Novel Model for Predicting PM2.5 Concentrations Utilizing Graph Convolutional Networks and Transformer
title_full_unstemmed A Novel Model for Predicting PM2.5 Concentrations Utilizing Graph Convolutional Networks and Transformer
title_short A Novel Model for Predicting PM2.5 Concentrations Utilizing Graph Convolutional Networks and Transformer
title_sort novel model for predicting pm2 5 concentrations utilizing graph convolutional networks and transformer
topic PM2.5 concentration prediction
spatiotemporal prediction
graph convolutional networks (GCN)
Transformer
url https://ieeexplore.ieee.org/document/10884736/
work_keys_str_mv AT yuanhuang anovelmodelforpredictingpm25concentrationsutilizinggraphconvolutionalnetworksandtransformer
AT feilonghan anovelmodelforpredictingpm25concentrationsutilizinggraphconvolutionalnetworksandtransformer
AT qimengfeng anovelmodelforpredictingpm25concentrationsutilizinggraphconvolutionalnetworksandtransformer
AT yuanhuang novelmodelforpredictingpm25concentrationsutilizinggraphconvolutionalnetworksandtransformer
AT feilonghan novelmodelforpredictingpm25concentrationsutilizinggraphconvolutionalnetworksandtransformer
AT qimengfeng novelmodelforpredictingpm25concentrationsutilizinggraphconvolutionalnetworksandtransformer